Identification and Quantification of Mixed Antibiotics in Aqueous Solution Through Machine Learning Analysis of SERS Spectra

Introduction: The accumulation of antibiotic residues in natural environments poses critical risks to public health and ecosystem stability due to their biochemical and physiological impacts. Ensuring water safety by preventing antibiotic contamination necessitates the development of rapid and reliable detection methods. Surface-enhanced Raman spectroscopy (SERS) offers high sensitivity and specificity in identifying small molecules but faces challenges in distinguishing individual antibiotics within complex mixtures.

Methods: This study introduces a deep-learning-based model to analyze SERS spectra for the efficient identification of antibiotics in mixtures and precise determination of their concentrations. A simulated water environment with residues of ciprofloxacin, doxycycline, and levofloxacin was prepared. A computational framework combining a convolutional neural network (CNN) for classification and a non-negative elastic network (NN-EN) for quantification was applied to the SERS spectra data.

Results: The CNN model achieved 98.68% accuracy in identifying individual antibiotics in mixtures. Additionally, the Shapley Additive exPlanations (SHAP) analysis highlighted the model's capacity to target specific spectral peaks, while the NN-EN model accurately quantified each antibiotic’s concentration within the mixtures.

Conclusion: The combined use of SERS with CNN and NN-EN models presents a promising solution for the rapid and accurate detection and quantification of antibiotic residues in water, potentially enhancing efforts to monitor and control aquatic contamination.